On Monday, Meta officially rolled out the open-source weights for its AI large model, Muse Glimmer. This model boasts a 30-billion parameter architecture and incorporates 4-bit quantization technology, facilitating local operation on consumer-grade GPU devices (e.g., MacBook M4/M5 Max, RTX 5090) that come with 24GB/32GB of video memory. Leveraging distillation techniques, the model is a compressed version of the larger Muse Spark model, preserving its performance while reducing hardware demands. In real-world tests, Muse Glimmer demonstrates seamless conversation and real-time interaction capabilities, with all computations being processed locally. When pitted against mainstream benchmarks, Muse Glimmer outperformed models of a similar scale, such as Gemma 4-31B and Qwen 3.6-27B. Meta also revealed its plans to release the weights of the more robust Muse Spark version in the upcoming weeks. Additionally, the company will set up a $1 billion fund to bolster U.S. data center communities. This strategic move signifies Meta's renewed commitment to its open-source approach, aiming to advance its vision of "personal hyperintelligence" by making AI more accessible and countering the closed-source models of its competitors.